Heterogeneity of Acquired Resistance to Anti-EGFR Monoclonal Antibodies in Patients with Metastatic Colorectal Cancer

This article is featured in Highlights of This Issue, p. 2375

In the last decade, uncovering the molecular and genetic bases of metastatic colorectal cancer (mCRC) has enabled personalized treatments that improved patients' survival and limited administration of costly and potentially toxic treatments to selected groups of patients. In particular, anti-EGFR therapy is nowadays approved for subjects with RAS wild-type tumors (1–3). In addition, BRAF mutations are commonly recognized as an additional biomarker of primary resistance to cetuximab and panitumumab (4). Even if still not validated, HER2 and MET gene amplification are less frequent alterations that are also associated with treatment resistance (5, 6).

Despite the implementation of biomarkers in the daily clinical practice, patients who initially respond almost invariably develop secondary resistance. The genetic and molecular landscape of secondary resistance is heterogeneous and partially overlaps with that of primary resistance (7). Several resistance mechanisms have been described so far, both in vitro and in retrospective patients' cohorts: these include KRAS/NRAS (8, 9), BRAF (10, 11), and EGFR ectodomain (9, 12, 13) mutations, as well as KRAS (8), HER2, and MET (5, 6, 14) amplifications.

However, the relative contribution and clinical significance of these alterations is far from being exhaustive due to several reasons: invasiveness of biopsy procedures and ethical issues that limit the number of feasible rebiopsies at progression; uncertain clinical applicability of data obtained from rebiopsies and liquid biopsies; small and heterogeneous published series mostly investigating one resistance biomarker at a time; poor applicability of advanced technologies in the real-life setting. In this scenario, a comprehensive characterization of acquired resistance mechanisms, their relative distribution, and co-occurrence in the same tumor or patient will hopefully help to define effective strategies to overcome or delay acquired resistance to treatment. Analysis of plasma circulating tumor DNA (ctDNA), if coupled with highly sensitive technologies, is able to detect mutant DNA alleles down to 0.01% (15) and has been proposed as a novel method for noninvasive detection of heterogeneous molecular alterations underlying the evolution of resistance in metastatic tumors (11, 16).

In this study, we extensively characterized the most common mechanisms of acquired resistance to anti-EGFR mAbs in patients with mCRC, with the aim of appreciating their significance in a real-life setting and correlating tissue- and blood-based data including allele frequency and molecular heterogeneity.

Eligible patients had diagnosis of mCRC and received cetuximab- or panitumumab-based therapy until progressive disease (PD) at Fondazione IRCCS Istituto Nazionale dei Tumori (Milan, Italy) and Istituto Oncologico Veneto of Padua (Padova, Italy). Since the detection of tumor clones harboring resistance mechanisms may significantly drop after treatment discontinuation (17), collection of tumor rebiopsy and blood samples was always performed within one month (30 days) from detection of PD.

Acquired resistance to anti-EGFR mAb, defined as radiographic documentation of PD after prior objective response according to RECIST 1.1 to single-agent panitumumab or irinotecan plus cetuximab in clearly irinotecan-refractory disease. Reintroduction of anti-EGFR monotherapy following a prior line of chemotherapy plus anti-EGFRs (discontinued for reasons other than PD) was allowed only if achieving a second RECIST response. At the time of PD documentation, tumor rebiopsy was performed on the most accessible site of progressing metastasis. Radiologic documentation of PD events preceding tumor rebiopsy should have occurred within 12 weeks from last dose of anti-EGFR therapy. Progression-free survival (PFS) and overall survival (OS) were calculated from the date of start of anti-EGFR treatment to the date of PD or death (or last follow up), respectively. All patients signed a written informed consent to allow rebiopsy procedures. Their consent was expressed within an agreement to participate to an observational prospective cohort study (INT 117/15) as part of a wide screening program for ethically approved, institutional phase I trials.

All archival tumor samples were RAS and BRAF wild-type as analyzed by means of Sanger sequencing and had to be reanalyzed with the same techniques used for postprogression tumor samples, as detailed in the next section. Only patients with tumors negative for both for MET and HER2 by silver in situ hybridization (SISH) and IHC were included. Pretreatment amplified clones had to be below the cutoff of 5% to be considered negative. Next-generation sequencing (NGS) had to confirm RAS and BRAF wild-type status and mutated subclones had to be undetectable.

On both archival and postprogression samples, we performed: (i) NGS of 50 genes' hotspot regions included in the Hotspot Cancer Panel v2 (Life Technologies) by using the Ion Torrent Personal Genome Machine platform (Life Technologies); (ii) dual color SISH and IHC both for MET and HER2. On postprogression samples only, we used a second custom panel to analyze 26 amplicons (2,77 kb) corresponding to the EGFR extracellular region (exons from 1 to 14), as described previously (18). Detailed description of the experimental methods is described in the Supplementary Materials.

Liquid biopsies were collected as previously described (17). Briefly, 10 mL of whole blood was obtained in EDTA tubes. Plasma was separated within 5 hours and stored at −80°C until ctDNA extraction by Maxwell RSC ccfDNA Plasma Kit with the automated Maxwell RSC Instrument (Promega) according to the manufacturer's instructions.

Droplet digital PCR (ddPCR) was performed as described previously [Siravegna and colleagues (2015) and Arena and colleagues (2016)]. The results were reported as fractional abundance of mutant DNA alleles to total (mutant plus wild-type) DNA alleles using the QuantaSoft analysis software (Bio-Rad) ddPCR analysis of normal control (from cell lines) and no DNA template controls were always included. Samples with too low positive events were repeated at least twice in independent experiments to validate the obtained results.

We analyzed samples from 22 prospectively treated patients, who achieved benefit from an anti-EGFR mAb, and then developed resistance. Their demographics and disease characteristics are listed in Table 1. All except one patient had left-sided tumors. Archival samples mostly came from the primary site (45.5%) or the liver (45.5%). Biopsies at disease progression were performed on liver (54.5%), lung (13.5%), primary tumor site (9%) or other metastatic sites (23% collectively). At the time of PD, all tissues were sampled by needle biopsy except for the case of patients 5, 14, and 15, where a whole metastatic lesion was surgically removed. Median time elapsed from PD to tumor rebiopsy was 20 (range 7–30) days.

Overall, 13 patients (59%) received panitumumab and 9 patients (41%) received cetuximab; half of the patients were treated with reintroduction of anti-EGFR monotherapy after previous discontinuation for reason other than PD.

Median PFS and OS were 8 and 15.6 months, respectively. No differences in PFS and OS were observed according to treatment status (reintroduction: yes vs. no).

Table 2 summarizes the genetic alterations found in tumor biopsies after progression on anti-EGFR mAbs therapy. Among 22 evaluated patients, KRAS mutations were found in 7 patients (2 Q61H, 1 Q61K, 2 G12R, 1 G13D, 1 G12V), BRAF mutations in 2 patients (both were V600E), HER2 amplifications in 3 patients, and, finally, MET amplifications in 4 patients. A potentially new resistance mechanism was found in patient no. 15 (Table 2), with a mutation in AKT1 gene, determining a D46N amino acid substitution, detected in 18% of analyzed cells. All DNA alterations found in the same biopsy sample were mutually exclusive, except for the case of patient no.16, in which both MET amplifications and KRAS Q61H mutation were found. No EGFR ectodomain mutation was detected.

In analyzed biopsies, the putative resistance mechanism was detected in a highly variable percentage of analyzed cells, independently from the specific mechanism or the detection technique (Table 2). In particular, median and average mutant allelic frequencies normalized for tumor cell content were 34% and 37.28%, respectively (range 6%–97%). The heterogeneity score (HS), an estimate of the percentage of cells expressing a mutant KRAS version (19), varied in the range of 12 to 110, with a median HS of 68 and an average HS of 62. This means that in some patients (e.g., patient 5), the mutant allele was present in only a minority of cells, while in patient 22, mutant KRAS was probably associated with gene amplification. The two cases with acquired BRAF mutations also presented low HSs (14 and 28). Moreover, HER2 amplification was found only in small cell fraction in patient 5 (Fig. 1), while heterogeneous MET amplification was described in patient 15 (Fig. 2); interestingly, these two cases with heterogeneous IHC staining pattern are among the three patients whose sampling consisted in complete metastasis removal and not needle biopsy, as previously explained. Finally, we found that patients with acquired MET amplification had a significantly shorter median PFS during anti-EGFRs as compared with those without (6.6 vs. 10 months, respectively; P = 0.014; Supplementary Fig. S1). On the other hand, neither KRAS/BRAF mutations nor HER2 amplification were predictive of shorter or longer PFS (data not shown).

For 11 of 22 patients included in our study, blood samples collected at the time of resistance were available and plasma ctDNA was successfully analysed by ddPCR. In particular, we looked for hotspot mutations in KRAS, NRAS, BRAF, and EGFR ectodomain gene sequences, as well as MET and HER2 copy number variation (CNV). We found 3 KRAS mutations, 1 EGFR ectodomain mutation, 1 concomitant KRAS mutation and HER2 amplification, 1 concomitant BRAF mutation and MET amplification. Most of gene mutations were found to have a low fractional abundance (ranging between 0.11% and 34%; median and average 7.83% and 12.67%, respectively), suggesting the presence of the mutation in a small fraction of DNA-releasing tumor cells.

Mutational and CNV analysis on ctDNA samples and their comparison with tissue-based data are summarized in Table 3 and easily assessable in Fig. 3. Among 11 blood samples available, ctDNA analysis was fully concordant with tissue biopsy analysis in 4 cases (patients no. 1, 7, 13, 22). Conversely, one patient with undetected resistance mechanisms in tissue biopsy had mutated KRAS and amplified HER2 in ctDNA (pt no. 14), while 3 patients with detected resistance mechanisms on biopsies had no alterations in liquid biopsies (pts no. 4, 5, 17), and, finally, in 3 cases with positive tissue biopsies, different or additional molecular alterations were detected in ctDNA (pts n. 12, 16, 20). Of note, 3 KRAS mutations with the highest HSs on tissue biopsies were confirmed by ctDNA analysis with relatively high fractional abundance. On the other hand, two cases of KRAS- (pt n.5) and BRAF- (pt n.17) mutated tumors with low HS on tissue sample analysis were not confirmed by ctDNA analyses. Finally, EGFR (pt no.12) and BRAF (pt no.20), but not KRAS (pt no. 14), mutations detected only in ctDNA were found to have a low fractional abundance (0.11%, 3.6%, and 21.5%, respectively). Collectively considered, these results suggest that mutations that are present in the tumor bulk of one biopsied lesion are usually detected with both methods, while alterations that are present in a minority of cells are more easily lost with either method.

To the best of our knowledge, this is the first study that extensively analyzed mechanisms of secondary resistance to anti-EGFR mAbs both in tissue and liquid biopsy analyses in patients with mCRC. As anti-EGFR mAbs are often combined with chemotherapy agents, it is difficult to discriminate between activity of chemotherapy and anti-EGFR agents in individual cases. To circumvent this problem and to optimize patients' selection on the basis of initial sensitivity to anti-EGFR therapy, we only included initially RAS-BRAF wild-type (20), HER2, and MET nonamplified patients with a radiologically documented objective response to anti-EGFR monotherapy (or in combination with irinotecan in clearly irinotecan-refractory disease). In our opinion, this combination of genetic and clinical criteria guarantees the best selection of patients previously benefiting from cetuximab/panitumumab sensitive tumors as possible. Interestingly, all but one tumors were left-sided, thus indirectly confirming recent evidences from randomized trials suggesting a different efficacy of anti-EGFRs based on primary tumor location (21, 22).

We found that the onset of secondary resistance to cetuximab or panitumumab is often associated with molecular alterations that likely converge on MAPK pathway reactivation, with KRAS mutations and MET or HER2 amplification being the most frequent, thus pointing out the potential clinical relevance of inhibiting the MAPK cascade in resistant tumors (10). However, KRAS-activating mutations, the most frequent mechanism, have not been successfully targeted so far (23); whether inhibitors of downstream KRAS effectors, for example, MEK or ERK1/2, can successfully target KRAS-mutated CRCs will be investigated in future studies (10). On the other hand, other emerging resistance mechanisms, including BRAF and PIK3CA mutations, as well as MET and HER2 amplifications, may be targeted by already available drugs or drug combinations, such as MEK1/2 or BRAF inhibitors combined with anti-EGFR mAbs in BRAF-mutated tumors (10, 24), the PIK3CA inhibitor alpelisib, the ALK/MET dual inhibitor crizotinib (18), and the HER2-targeted dual block with trastuzumab and lapatinib or trastuzumab and pertuzumab (25, 26). The very low rate (1/22 patients) of mutations in PIK3CA/AKT/mTOR pathway indicates that, differently from metastatic melanoma developing acquired resistance to BRAF inhibitors (27), this pathway is a less crucial determinant of tumor resistance to cetuximab or panitumumab in mCRCs. Even if codon 46 AKT1 mutations were already described as somatic variants (28, 29), their functional validation should be assessed by future preclinical studies. In this study, we were able to detect EGFR ectodomain mutations in only 1 of 11 ctDNA available samples, and also with low fractional abundance; this suggests that this resistance mechanism is rare overall, and occurs in a small proportion of tumor cells.

We also found that the emergence of MET amplification correlates with poorer PFS during anti-EGFR treatment. This mechanism was already shown for RAS mutations as compared with EGFR ectodomain mutations (30). We have also previously demonstrated that pre-existing MET-amplified subclones drive tumor progression; now, we suggest that the aggressiveness of such emerging clones may be associated with shorter benefit from anti-EGFR agents (14, 18).

In postprogression mutational analyses, we mostly observed a variable proportion, and rarely the totality, of sampled cells expressing one specific resistance mechanism. This was especially evident in the case of KRAS-mutated tumors, where a highly variable HS was calculated. The most intuitive interpretation to this finding is that different cell subpopulations within a single metastasis may have selected different resistance mechanisms, and we only detected those alterations that have been reported in the literature (intralesion heterogeneity). One alternative hypothesis is that nonmutated subpopulations consist of nonresistant cells in dynamic equilibrium with resistant ones. However, this latter interpretation is highly unlikely, because resistant cells rapidly replace sensitive cells in progressing lesions (7), while we found in some cases only a small proportion (< 10%) of cells harbouring one specific mechanism of resistance. Of note, the finding of heterogeneous HER2 and MET amplification in two patients (patient 5 and 15, respectively) whose sampling at progression derived from surgical removal of their whole metastases instead of needle biopsy, may indicate that more accurate sampling can improve detection of small clusters of resistant cells that are embedded within the tumor bulk. In our opinion, these data coherently support the hypothesis of intralesion heterogeneity in resistant tumors (19), as also suggested by the finding of two resistance mechanisms in different cell proportions within the same biopsy specimen (patient no. 16).

By a clinical perspective, it is currently unknown whether and how the accumulating evidence on mechanisms of acquired resistance to anti-EGFR mAbs will translate into palpable clinical benefit in patients with mCRC. The evidence of intralesion and interlesion heterogeneity in resistant mechanisms would likely preclude the possibility to successfully target all resistant clones with a single targeted approach. However, case reports published so far suggest that, despite clonal heterogeneity, one mechanism may become prevalent in some patients, who could therefore benefit from targeted inhibition of the selected mechanism (18, 31). However, our results underline how hardly mCRC will replicate recent successes in the treatment of EGFR-mutated lung adenocarcinomas becoming resistant to first-generation TKIs, in which third-generation TKIs targeting T790M-mutated EGFR are able to produce prolonged disease remissions or stabilizations (32, 33).

To our surprise, in this study, we observed poor concordance between DNA analysis on tissue biopsy samples and ctDNA compared with previously reported data (17). In particular, in 3 of 11 cases, high-sensitivity ddPCR analysis of ctDNA did not confirm results found in tissue DNA analysis. One possible explanation is that, compared with pretreatment conditions (11, 17), tumors resistant to anti-EGFR blockade display significantly higher genetic heterogeneity, with several cell clones harboring different genetic alterations; this may cause the coexistence of several resistant clones in different metastatic lesions, with the chance to miss the opportunity to catch some of them with either tissue or liquid biopsy. Our data also suggest that results of genetic analyses may be deeply affected by the sampling method and/or DNA sequencing techniques employed; in some cases, high-sensitivity ddPCR analysis of ctDNA can help to identify alterations that are not revealed in tissue biopsies, due to their low fractional abundance (as is probably the case of patient no. 12), or their emergence only in nonsampled lesions. On the other hand, genetic alterations found on tissue, but not liquid biopsies, may indicate heterogeneity of DNA release by different tumor cell subpopulations, probably as a consequence of clone-related (clone genetics, apoptotic rate) and environment-related variables (e.g., tumor vascularization).

Therefore, the best way to comprehensively characterize acquired resistance is far from being fully established. Despite the limits related to procedure invasiveness and limited sampling, tissue biopsies are less influenced by tumor environment characteristics, and also provide quantitative information (adjustment of minor allele frequency according to tumor cellularity) on the prevalence of the resistance mechanism(s) within the bulk of tumor cells. On the other hand, it is well known that single-lesion biopsies are often missing to capture tumor heterogeneity (31). In the future, combining tissue biopsy, ctDNA analysis, and radiologic imaging could help to fully characterize the heterogeneity of occurring resistance mechanisms, while also clarifying their relative contribution to the clinical resistance.

In this study, mechanisms of acquired resistance remain uncharacterized for the 23% of evaluated patients. However, due to the presence of intra- and interlesion heterogeneity, this percentage is likely to be an underestimation of the real number of tumor subpopulations with still uncharacterized resistance mechanisms. This should stimulate future research to identify novel mechanisms of secondary resistance to anti-EGFR mAbs.

In conclusion, this study highlights MAPK reactivation as the most prominent mechanism driving secondary resistance to cetuximab and panitumumab in metastatic CRCs. The fact that different pathways can converge on MAPK cascade reactivation excludes upfront cotargeting of EGFR and kinases upstream of these pathways to prevent selection of resistant clones, because cells would rapidly evolve and/or select alternative escape mechanisms. However, early detection of emerging mechanisms of resistance through liquid biopsies may precociously suggest rational combinations of anti-EGFR mAbs with agents targeting the evolving resistance mechanism (16).

A. Bardelli is a consultant/advisory board member for Biocartis, Horizon Discovery, and Trovagene. No potential conflicts of interest were disclosed by the other authors.

Conception and design: F. Pietrantonio, A. Mennitto, M. Milione, F. de Braud

Development of methodology: F. Pietrantonio, G. Siravegna, F. Perrone, E. Tamborini, C.C. Volpi

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F. Pietrantonio, G. Siravegna, A. Mennitto, R. Berenato, S. Lonardi, F. Morano, A. Martinetti, F. Battaglin, I. Bossi, A. Pellegrinelli, C. Cremolini, M. Di Bartolomeo

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F. Pietrantonio, C. Vernieri, G. Siravegna, R. Berenato, F. Perrone, A. Gloghini, E. Tamborini, S. Lonardi, F. Morano, B. Picciani, A. Busico, A. Martinetti, F. Battaglin, A. Pellegrinelli, C. Cremolini, M. Di Bartolomeo, A. Bardelli

Writing, review, and/or revision of the manuscript: F. Pietrantonio, C. Vernieri, G. Siravegna, A. Mennitto, R. Berenato, A. Gloghini, S. Lonardi, F. Morano, C.C. Volpi, F. Battaglin, C. Cremolini, M. Di Bartolomeo, A. Bardelli, F. de Braud

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): F. Pietrantonio, I. Bossi, M.D. Bartolomeo

Study supervision: F. Pietrantonio, M. Di Bartolomeo, A. Bardelli, F. de Braud

This work was supported by European Community's Seventh Framework Programme under grant agreement no. 602901 MErCuRIC (to A. Bardelli); H2020 no. 635342-2 MoTriColor (to A. Bardelli); IMI contract no. 115749 CANCER-ID (to A. Bardelli); AIRC 2010 Special Program Molecular Clinical Oncology 5 per mille, Project no. 9970 (to A. Bardelli); AIRC IG no. 16788 (to A. Bardelli); Fondazione Piemontese per la Ricerca sul Cancro-ONLUS 5 per mille 2011 Ministero della Salute (to A. Bardelli).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.